Input Error Messages

[PDoakORNL]

Currently DCA does little checking of user input for validity or match with the built executable model etc. This needs to improve even for expert users.

1. If working directory is not present hdf5 output crashes with vague error.
2. Model sections for models other than the one the executable is built for are silently ignored.
3. Model sections with name typos are silently ignored and model is run with defaults.
4. Numeric inputs are not checkedd to be numeric data resulting in crashes due to random numerical errors later.

[tylersax]

@PDoakORNL for problem 2 described above, is "multiple model sections in the input file" something to throw on, or just warn?

I see test cases in the code that have multiple model sections and would fail if we hard throw any time we read multiple model sections (e.g. test/unit/pays/dca_step/cluster_solver/shared_tools/accumulation/tp/input_1x1_multi.json ). I see two options (but please add a third if I'm missing something):

1. Having multiple model sections is a legitimate use case (say the user wants to sweep across parameters/models, but otherwise keep the same input file). The program should warn at runtime, but proceed.
2. Having multiple model sections is a bad pattern. We should throw, forcing the user to narrow down to one. This route will require splitting up test cases that have multiple model sections.

[tylersax]

Per offline discussion, we are going to pursue approach 1. That is, we will warn in the case where extra model sections are included but not built against.

Relatedly, we discussed the issue of input files that have no model parameters at all. We will treat this as similar to bug 3 described above, and will throw in these cases since it is very unlikely that the user intends to run DCA++ with no model parameters.